探索查找异常数据¶
In [1]:
import os
os.chdir("{}/image".format(os.getcwd()))
I. 获取训练集中高度或者宽度小于50的图片。¶
In [2]:
from keras.preprocessing.image import img_to_array, load_img
import os
import numpy as np
import matplotlib.pyplot as plt
targetnames = os.listdir('train/')
bad_picture = []
for name in targetnames[1:]:
img = load_img('train/'+name)
x = img_to_array(img)
if x.shape[0] < 50 or x.shape[1] < 50:
bad_picture.append(name)
print(bad_picture)
In [3]:
import matplotlib.pyplot as plt
from keras.preprocessing.image import load_img
# from keras.preprocessing import image
from math import ceil
plt.style.use('seaborn-white')
def show_img_list(img_list, size = (12, 12)):
print("Len img_list: {}".format(len(img_list)))
plt.figure(figsize=size)
subplot_row = ceil(len(img_list) / 5)
for i in range(0, len(img_list)):
plt.subplot(subplot_row, 5, i+1)
img = load_img('train/'+ img_list[i])
plt.title(img_list[i])
plt.axis('off')
plt.imshow(img)
show_img_list(bad_picture)
结论，非猫狗异常图片：dog.10747.jpg, cat.9171.jpg, dog.4367.jpg¶
II. 通过对图片中的色彩-像素比进行 IQR 分析(可发现很多分辨率低、无关的图片)¶
In [4]:
from PIL import Image
import os
import numpy as np
import shutil
from collections import Counter
targetnames = os.listdir('train/')
ratio_list = []
for name in targetnames[1:]:
im = Image.open('train/' + name)
x = im.histogram(mask=None)
count = Counter(x)
ratio_list.append(float(len(count))/len(x))
In [5]:
import numpy as np
q99, q01 = np.percentile(ratio_list, [99, 1])
print(q99, q01)
In [6]:
from keras.preprocessing.image import img_to_array, load_img
import shutil
import os
import matplotlib.pyplot as plt
from math import ceil
%matplotlib inline
plt.style.use('seaborn-white')
outlier_picture = []
targetnames = os.listdir('train/')
for name in targetnames[:]:
im = Image.open('train/' + name)
x = im.histogram(mask=None)
count = Counter(x)
if float(len(count))/len(x) < q01:
outlier_picture.append(name)
img = load_img('train/'+name)
plt.title(name)
plt.imshow(img)
plt.show()
print(outlier_picture)
结论，非猫狗异常图片：cat.8456.jpg, dog.5604.jpg, dog.10747.jpg, dog.8736.jpg, cat.9171.jpg, dog.9517.jpg, cat.11184.jpg, dog.4367.jpg¶
III. train数据集，采用预处理模型查找最可能不是猫狗的图片¶
In [7]:
import csv
import matplotlib.pyplot as plt
%matplotlib inline
plt.style.use('seaborn-white')
def get_imageNet_class(file_path):
category_class = []
with open(file_path, 'r') as f:
reader = csv.reader(f)
for line in reader:
if line[1] == '狗' or line[1] == '猫':
category_class.append(line[0])
return category_class
imageNet_class = get_imageNet_class('ImageNetClasses.csv')
In [8]:
from keras.models import *
from keras.layers import *
from keras.applications import *
from keras.preprocessing import image
from keras.applications.resnet50 import preprocess_input, decode_predictions
from keras.applications.inception_v3 import preprocess_input, decode_predictions
from keras.applications.xception import preprocess_input, decode_predictions
from keras.applications.inception_resnet_v2 import InceptionResNetV2,preprocess_input,decode_predictions
import numpy as np
import random
from math import ceil
def preprocess_input(x):
return x - [103.939, 116.779, 123.68]
def get_outlier_predictions(MODEL, img_size, preprocess_input, decode_predictions):
model = MODEL(weights='imagenet')
outlier_predictions = {}
targetnames = os.listdir('train/')
for name in targetnames[:]:
img = image.load_img('train/' + name, target_size=img_size)
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
preds = model.predict(x)
predictions = decode_predictions(preds, top=10)[0]
#if predictions not in imageNet_class:
# outlier_predictions.append(name)
outlier_predictions[name] = predictions
return outlier_predictions
In [9]:
import csv
def writer_csv(csv_path, data):
"""
:type csv_path: str
:type data: dict
:rtype: None
"""
with open(csv_path, 'w') as csv_file:
writer = csv.writer(csv_file)
for key, value in data.items():
writer.writerow([key, value])
def readr_csv(csv_path):
"""
:type cav_path: str
:rtpye: dict
"""
with open(csv_path, 'r') as csv_file:
reader = csv.reader(csv_file)
return dict(reader)
def get_result_list(pred):
result = []
for key, value in pred.items():
value = eval(value)
pred_list = [ x[0] for x in value ]
pred_list = [ 1 for x in pred_list if x in imageNet_class ]
if sum(pred_list) == 0:
result.append(key)
return result
In [10]:
import cv2
def plt_outlier_img_1(outlier_list, img_size):
plt.figure(figsize=(16, 68), dpi=80)
subplot_row = ceil(len(outlier_list) / 5)
for i in range(0, len(outlier_list)):
plt.subplot(subplot_row, 5, i+1)
img = cv2.imread('train/'+ outlier_list[i])
img = cv2.resize(img, img_size)
x = img.copy()
x.astype(np.float32)
plt.title(outlier_list[i])
plt.axis('off')
plt.imshow(x[:,:,::-1])
# img = image.load_img('train/'+ outlier_list[i])
# x = image.img_to_array(img)
# plt.title(outlier_list[i])
# #plt.axis('off')
# plt.tight_layout()
# plt.imshow(img)
In [11]:
csv_path = 'train_decode_predictions.csv'
In [12]:
def get_outlier_result(MODEL, img_size, preprocess_input, decode_predictions):
outlier_list = get_outlier_predictions(MODEL, img_size, preprocess_input, decode_predictions)
writer_csv(csv_path, outlier_list)
outlier_list = readr_csv(csv_path)
len(outlier_list)
outlier_result = get_result_list(outlier_list)
len(outlier_result)
return outlier_result
# targetnames = os.listdir('image/train/')
In [13]:
def get_counts_cat_dog(result_list):
n = len(result_list)
cat = 0
dog = 0
for filename in result_list:
name = filename[0: 3]
if(name == 'cat'):
cat += 1
elif(name == 'dog'):
dog += 1
else:
return
print('all = {}, cat = {}, dog = {}'.format(n, cat, dog))
return n, cat, dog
注意： 运行下面的代码，下载预训练模型可能会网络超时，例如：Exception: URL fetch failure on https://github.com/fchollet/deep-learning-models/releases/download/v0.4/xception_weights_tf_dim_ordering_tf_kernels_notop.h5: None -- [Errno 110] Connection timed out
此时，需要手动下载先对应的模型文件，放到~/.keras/models/ 文件路径下，即可。
In [14]:
resnet50_result = get_outlier_result(ResNet50, (224, 224), resnet50.preprocess_input, resnet50.decode_predictions)
get_counts_cat_dog(resnet50_result)
plt_outlier_img_1(resnet50_result, (224, 224))
异常图片为： ['dog.1773.jpg', 'dog.6475.jpg', 'dog.12376.jpg','dog.10237.jpg', 'cat.10712.jpg', 'dog.10747.jpg','dog.11299.jpg', 'cat.8456.jpg', 'dog.5604.jpg', 'dog.8736.jpg', 'cat.9171.jpg', 'cat.7564.jpg', 'dog.9517.jpg', 'dog.2614.jpg', 'dog.1043.jpg', 'cat.5351.jpg', 'dog.10801.jpg', 'cat.7377.jpg', 'cat.5418.jpg', 'cat.4338.jpg', 'cat.11184.jpg','dog.4367.jpg']
背景复杂，可疑异常图片为: ['cat.10029.jpg', 'dog.9418.jpg', 'cat.12272.jpg', 'dog.1835.jpg', 'cat.3868.jpg','cat.10636.jpg', 'cat.3731.jpg', 'dog.4507.jpg', 'dog.10123.jpg','cat.10700.jpg', 'dog.1161.jpg', 'dog.3889.jpg', 'cat.4308.jpg', 'dog.59.jpg', 'dog.6725.jpg', 'dog.7727.jpg', 'dog.10161.jpg', 'dog.9681.jpg', 'cat.7372.jpg', 'dog.11437.jpg', 'dog.6405.jpg', 'cat.9520.jpg', 'dog.3994.jpg','cat.7550.jpg', 'cat.3672.jpg', 'dog.4218.jpg']
In [16]:
xception_result = get_outlier_result(Xception, (299, 299), xception.preprocess_input, xception.decode_predictions)
get_counts_cat_dog(xception_result)
plt_outlier_img_1(xception_result, (299, 299))
异常图片为：
['dog.1773.jpg', 'dog.6475.jpg', 'dog.10237.jpg', 'dog.12376.jpg', 'cat.10712.jpg', 'dog.10747.jpg', 'dog.11299.jpg', 'cat.8456.jpg', 'dog.5604.jpg', 'dog.8736.jpg', 'cat.9171.jpg', 'cat.7564.jpg', 'dog.9517.jpg', 'dog.2614.jpg', 'dog.1043.jpg', 'cat.5351.jpg', 'dog.10801.jpg','cat.7377.jpg', 'cat.5418.jpg','cat.4338.jpg', 'cat.11184.jpg','dog.4367.jpg']
In [17]:
inceptionV3_result = get_outlier_result(InceptionV3, (299, 299), inception_v3.preprocess_input, inception_v3.decode_predictions)
get_counts_cat_dog(inceptionV3_result)
plt_outlier_img_1(inceptionV3_result, (299, 299))
